They really, really wanted to become an "enterprise" company. They acquired certain management people who saw development tools as a dead end, and decided all the money was in the enterprise.

But they misjudged the market, and invested in products like Visibroker and Midas (and dead-end tech like CORBA), not really understanding what they ought to be building.

(This wasn't the beginning of their troubles, though. They bungled a lot of decisions before then, especially with the purchase of dBase and Paradox, and the Windows version of Quattro Pro.)

I stopped using Delphi in the late 1990s, but even then I could see how it was heading towards obsolescence. Not the GUI development model itself, which was brilliant and which I sorely miss for OS X GUI development today, but Object Pascal.

Sure, Object Pascal is a good language, but it was very insular, not portable, often very Windows-specific, didn't integrate well with other things, and didn't evolve fast enough. I was able to work with C and C++ libraries, including Dialogic hardware APIs and Microsoft tech such as MAPI and TAPI, by writing a translator (htrans) that parsed header files into an AST and produced Object Pascal interface files for them. But it was very tiring. I had high hopes for C++Builder, but it was a major disappointment when it arrived. For one, it relied on a lot of proprietary C++ extensions.

I was using Delphi/Object Pascal for a lot a systems development, non-GUI backend stuff. It seems a bit quixotic today, but it was fun, and I liked the language. It sort of filled a particular niche that Java would overtake a few years later.